kernel/panic.c

   1 /*
   2  *  linux/kernel/panic.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * This function is used through-out the kernel (including mm and fs)
   9  * to indicate a major problem.
  10  */
  11 #include <linux/debug_locks.h>
  12 #include <linux/interrupt.h>
  13 #include <linux/kmsg_dump.h>
  14 #include <linux/kallsyms.h>
  15 #include <linux/notifier.h>
  16 #include <linux/module.h>
  17 #include <linux/random.h>
  18 #include <linux/ftrace.h>
  19 #include <linux/reboot.h>
  20 #include <linux/delay.h>
  21 #include <linux/kexec.h>
  22 #include <linux/sched.h>
  23 #include <linux/sysrq.h>
  24 #include <linux/init.h>
  25 #include <linux/nmi.h>
  26 #include <linux/console.h>
  27
  28 #define PANIC_TIMER_STEP 100
  29 #define PANIC_BLINK_SPD 18
  30
  31 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
  32 static unsigned long tainted_mask;
  33 static int pause_on_oops;
  34 static int pause_on_oops_flag;
  35 static DEFINE_SPINLOCK(pause_on_oops_lock);
  36 bool crash_kexec_post_notifiers;
  37 int panic_on_warn __read_mostly;
  38
  39 int panic_timeout = CONFIG_PANIC_TIMEOUT;
  40 EXPORT_SYMBOL_GPL(panic_timeout);
  41
  42 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
  43
  44 EXPORT_SYMBOL(panic_notifier_list);
  45
  46 static long no_blink(int state)
  47 {
  48         return 0;
  49 }
  50
  51 /* Returns how long it waited in ms */
  52 long (*panic_blink)(int state);
  53 EXPORT_SYMBOL(panic_blink);
  54
  55 /*
  56  * Stop ourself in panic -- architecture code may override this
  57  */
  58 void __weak panic_smp_self_stop(void)
  59 {
  60         while (1)
  61                 cpu_relax();
  62 }
  63
  64 /*
  65  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
  66  * may override this to prepare for crash dumping, e.g. save regs info.
  67  */
  68 void __weak nmi_panic_self_stop(struct pt_regs *regs)
  69 {
  70         panic_smp_self_stop();
  71 }
  72
  73 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
  74
  75 /**
  76  *      panic - halt the system
  77  *      @fmt: The text string to print
  78  *
  79  *      Display a message, then perform cleanups.
  80  *
  81  *      This function never returns.
  82  */
  83 void panic(const char *fmt, ...)
  84 {
  85         static char buf[1024];
  86         va_list args;
  87         long i, i_next = 0;
  88         int state = 0;
  89         int old_cpu, this_cpu;
  90
  91         /*
  92          * Disable local interrupts. This will prevent panic_smp_self_stop
  93          * from deadlocking the first cpu that invokes the panic, since
  94          * there is nothing to prevent an interrupt handler (that runs
  95          * after setting panic_cpu) from invoking panic() again.
  96          */
  97         local_irq_disable();
  98
  99         /*
 100          * It's possible to come here directly from a panic-assertion and
 101          * not have preempt disabled. Some functions called from here want
 102          * preempt to be disabled. No point enabling it later though...
 103          *
 104          * Only one CPU is allowed to execute the panic code from here. For
 105          * multiple parallel invocations of panic, all other CPUs either
 106          * stop themself or will wait until they are stopped by the 1st CPU
 107          * with smp_send_stop().
 108          *
 109          * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
 110          * comes here, so go ahead.
 111          * `old_cpu == this_cpu' means we came from nmi_panic() which sets
 112          * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
 113          */
 114         this_cpu = raw_smp_processor_id();
 115         old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
 116
 117         if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
 118                 panic_smp_self_stop();
 119
 120         console_verbose();
 121         bust_spinlocks(1);
 122         va_start(args, fmt);
 123         vsnprintf(buf, sizeof(buf), fmt, args);
 124         va_end(args);
 125         pr_emerg("Kernel panic - not syncing: %s\n", buf);
 126 #ifdef CONFIG_DEBUG_BUGVERBOSE
 127         /*
 128          * Avoid nested stack-dumping if a panic occurs during oops processing
 129          */
 130         if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
 131                 dump_stack();
 132 #endif
 133
 134         /*
 135          * If we have crashed and we have a crash kernel loaded let it handle
 136          * everything else.
 137          * If we want to run this after calling panic_notifiers, pass
 138          * the "crash_kexec_post_notifiers" option to the kernel.
 139          *
 140          * Bypass the panic_cpu check and call __crash_kexec directly.
 141          */
 142         if (!crash_kexec_post_notifiers)
 143                 __crash_kexec(NULL);
 144
 145         /*
 146          * Note smp_send_stop is the usual smp shutdown function, which
 147          * unfortunately means it may not be hardened to work in a panic
 148          * situation.
 149          */
 150         smp_send_stop();
 151
 152         /*
 153          * Run any panic handlers, including those that might need to
 154          * add information to the kmsg dump output.
 155          */
 156         atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
 157
 158         kmsg_dump(KMSG_DUMP_PANIC);
 159
 160         /*
 161          * If you doubt kdump always works fine in any situation,
 162          * "crash_kexec_post_notifiers" offers you a chance to run
 163          * panic_notifiers and dumping kmsg before kdump.
 164          * Note: since some panic_notifiers can make crashed kernel
 165          * more unstable, it can increase risks of the kdump failure too.
 166          *
 167          * Bypass the panic_cpu check and call __crash_kexec directly.
 168          */
 169         if (crash_kexec_post_notifiers)
 170                 __crash_kexec(NULL);
 171
 172         bust_spinlocks(0);
 173
 174         /*
 175          * We may have ended up stopping the CPU holding the lock (in
 176          * smp_send_stop()) while still having some valuable data in the console
 177          * buffer.  Try to acquire the lock then release it regardless of the
 178          * result.  The release will also print the buffers out.  Locks debug
 179          * should be disabled to avoid reporting bad unlock balance when
 180          * panic() is not being callled from OOPS.
 181          */
 182         debug_locks_off();
 183         console_flush_on_panic();
 184
 185         if (!panic_blink)
 186                 panic_blink = no_blink;
 187
 188         if (panic_timeout > 0) {
 189                 /*
 190                  * Delay timeout seconds before rebooting the machine.
 191                  * We can't use the "normal" timers since we just panicked.
 192                  */
 193                 pr_emerg("Rebooting in %d seconds..", panic_timeout);
 194
 195                 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
 196                         touch_nmi_watchdog();
 197                         if (i >= i_next) {
 198                                 i += panic_blink(state ^= 1);
 199                                 i_next = i + 3600 / PANIC_BLINK_SPD;
 200                         }
 201                         mdelay(PANIC_TIMER_STEP);
 202                 }
 203         }
 204         if (panic_timeout != 0) {
 205                 /*
 206                  * This will not be a clean reboot, with everything
 207                  * shutting down.  But if there is a chance of
 208                  * rebooting the system it will be rebooted.
 209                  */
 210                 emergency_restart();
 211         }
 212 #ifdef __sparc__
 213         {
 214                 extern int stop_a_enabled;
 215                 /* Make sure the user can actually press Stop-A (L1-A) */
 216                 stop_a_enabled = 1;
 217                 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
 218         }
 219 #endif
 220 #if defined(CONFIG_S390)
 221         {
 222                 unsigned long caller;
 223
 224                 caller = (unsigned long)__builtin_return_address(0);
 225                 disabled_wait(caller);
 226         }
 227 #endif
 228         pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
 229         local_irq_enable();
 230         for (i = 0; ; i += PANIC_TIMER_STEP) {
 231                 touch_softlockup_watchdog();
 232                 if (i >= i_next) {
 233                         i += panic_blink(state ^= 1);
 234                         i_next = i + 3600 / PANIC_BLINK_SPD;
 235                 }
 236                 mdelay(PANIC_TIMER_STEP);
 237         }
 238 }
 239
 240 EXPORT_SYMBOL(panic);
 241
 242
 243 struct tnt {
 244         u8      bit;
 245         char    true;
 246         char    false;
 247 };
 248
 249 static const struct tnt tnts[] = {
 250         { TAINT_PROPRIETARY_MODULE,     'P', 'G' },
 251         { TAINT_FORCED_MODULE,          'F', ' ' },
 252         { TAINT_CPU_OUT_OF_SPEC,        'S', ' ' },
 253         { TAINT_FORCED_RMMOD,           'R', ' ' },
 254         { TAINT_MACHINE_CHECK,          'M', ' ' },
 255         { TAINT_BAD_PAGE,               'B', ' ' },
 256         { TAINT_USER,                   'U', ' ' },
 257         { TAINT_DIE,                    'D', ' ' },
 258         { TAINT_OVERRIDDEN_ACPI_TABLE,  'A', ' ' },
 259         { TAINT_WARN,                   'W', ' ' },
 260         { TAINT_CRAP,                   'C', ' ' },
 261         { TAINT_FIRMWARE_WORKAROUND,    'I', ' ' },
 262         { TAINT_OOT_MODULE,             'O', ' ' },
 263         { TAINT_UNSIGNED_MODULE,        'E', ' ' },
 264         { TAINT_SOFTLOCKUP,             'L', ' ' },
 265         { TAINT_LIVEPATCH,              'K', ' ' },
 266 };
 267
 268 /**
 269  *      print_tainted - return a string to represent the kernel taint state.
 270  *
 271  *  'P' - Proprietary module has been loaded.
 272  *  'F' - Module has been forcibly loaded.
 273  *  'S' - SMP with CPUs not designed for SMP.
 274  *  'R' - User forced a module unload.
 275  *  'M' - System experienced a machine check exception.
 276  *  'B' - System has hit bad_page.
 277  *  'U' - Userspace-defined naughtiness.
 278  *  'D' - Kernel has oopsed before
 279  *  'A' - ACPI table overridden.
 280  *  'W' - Taint on warning.
 281  *  'C' - modules from drivers/staging are loaded.
 282  *  'I' - Working around severe firmware bug.
 283  *  'O' - Out-of-tree module has been loaded.
 284  *  'E' - Unsigned module has been loaded.
 285  *  'L' - A soft lockup has previously occurred.
 286  *  'K' - Kernel has been live patched.
 287  *
 288  *      The string is overwritten by the next call to print_tainted().
 289  */
 290 const char *print_tainted(void)
 291 {
 292         static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
 293
 294         if (tainted_mask) {
 295                 char *s;
 296                 int i;
 297
 298                 s = buf + sprintf(buf, "Tainted: ");
 299                 for (i = 0; i < ARRAY_SIZE(tnts); i++) {
 300                         const struct tnt *t = &tnts[i];
 301                         *s++ = test_bit(t->bit, &tainted_mask) ?
 302                                         t->true : t->false;
 303                 }
 304                 *s = 0;
 305         } else
 306                 snprintf(buf, sizeof(buf), "Not tainted");
 307
 308         return buf;
 309 }
 310
 311 int test_taint(unsigned flag)
 312 {
 313         return test_bit(flag, &tainted_mask);
 314 }
 315 EXPORT_SYMBOL(test_taint);
 316
 317 unsigned long get_taint(void)
 318 {
 319         return tainted_mask;
 320 }
 321
 322 /**
 323  * add_taint: add a taint flag if not already set.
 324  * @flag: one of the TAINT_* constants.
 325  * @lockdep_ok: whether lock debugging is still OK.
 326  *
 327  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
 328  * some notewortht-but-not-corrupting cases, it can be set to true.
 329  */
 330 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
 331 {
 332         if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
 333                 pr_warn("Disabling lock debugging due to kernel taint\n");
 334
 335         set_bit(flag, &tainted_mask);
 336 }
 337 EXPORT_SYMBOL(add_taint);
 338
 339 static void spin_msec(int msecs)
 340 {
 341         int i;
 342
 343         for (i = 0; i < msecs; i++) {
 344                 touch_nmi_watchdog();
 345                 mdelay(1);
 346         }
 347 }
 348
 349 /*
 350  * It just happens that oops_enter() and oops_exit() are identically
 351  * implemented...
 352  */
 353 static void do_oops_enter_exit(void)
 354 {
 355         unsigned long flags;
 356         static int spin_counter;
 357
 358         if (!pause_on_oops)
 359                 return;
 360
 361         spin_lock_irqsave(&pause_on_oops_lock, flags);
 362         if (pause_on_oops_flag == 0) {
 363                 /* This CPU may now print the oops message */
 364                 pause_on_oops_flag = 1;
 365         } else {
 366                 /* We need to stall this CPU */
 367                 if (!spin_counter) {
 368                         /* This CPU gets to do the counting */
 369                         spin_counter = pause_on_oops;
 370                         do {
 371                                 spin_unlock(&pause_on_oops_lock);
 372                                 spin_msec(MSEC_PER_SEC);
 373                                 spin_lock(&pause_on_oops_lock);
 374                         } while (--spin_counter);
 375                         pause_on_oops_flag = 0;
 376                 } else {
 377                         /* This CPU waits for a different one */
 378                         while (spin_counter) {
 379                                 spin_unlock(&pause_on_oops_lock);
 380                                 spin_msec(1);
 381                                 spin_lock(&pause_on_oops_lock);
 382                         }
 383                 }
 384         }
 385         spin_unlock_irqrestore(&pause_on_oops_lock, flags);
 386 }
 387
 388 /*
 389  * Return true if the calling CPU is allowed to print oops-related info.
 390  * This is a bit racy..
 391  */
 392 int oops_may_print(void)
 393 {
 394         return pause_on_oops_flag == 0;
 395 }
 396
 397 /*
 398  * Called when the architecture enters its oops handler, before it prints
 399  * anything.  If this is the first CPU to oops, and it's oopsing the first
 400  * time then let it proceed.
 401  *
 402  * This is all enabled by the pause_on_oops kernel boot option.  We do all
 403  * this to ensure that oopses don't scroll off the screen.  It has the
 404  * side-effect of preventing later-oopsing CPUs from mucking up the display,
 405  * too.
 406  *
 407  * It turns out that the CPU which is allowed to print ends up pausing for
 408  * the right duration, whereas all the other CPUs pause for twice as long:
 409  * once in oops_enter(), once in oops_exit().
 410  */
 411 void oops_enter(void)
 412 {
 413         tracing_off();
 414         /* can't trust the integrity of the kernel anymore: */
 415         debug_locks_off();
 416         do_oops_enter_exit();
 417 }
 418
 419 /*
 420  * 64-bit random ID for oopses:
 421  */
 422 static u64 oops_id;
 423
 424 static int init_oops_id(void)
 425 {
 426         if (!oops_id)
 427                 get_random_bytes(&oops_id, sizeof(oops_id));
 428         else
 429                 oops_id++;
 430
 431         return 0;
 432 }
 433 late_initcall(init_oops_id);
 434
 435 void print_oops_end_marker(void)
 436 {
 437         init_oops_id();
 438         pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
 439 }
 440
 441 /*
 442  * Called when the architecture exits its oops handler, after printing
 443  * everything.
 444  */
 445 void oops_exit(void)
 446 {
 447         do_oops_enter_exit();
 448         print_oops_end_marker();
 449         kmsg_dump(KMSG_DUMP_OOPS);
 450 }
 451
 452 #ifdef WANT_WARN_ON_SLOWPATH
 453 struct slowpath_args {
 454         const char *fmt;
 455         va_list args;
 456 };
 457
 458 static void warn_slowpath_common(const char *file, int line, void *caller,
 459                                  unsigned taint, struct slowpath_args *args)
 460 {
 461         disable_trace_on_warning();
 462
 463         pr_warn("------------[ cut here ]------------\n");
 464         pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS()\n",
 465                 raw_smp_processor_id(), current->pid, file, line, caller);
 466
 467         if (args)
 468                 vprintk(args->fmt, args->args);
 469
 470         if (panic_on_warn) {
 471                 /*
 472                  * This thread may hit another WARN() in the panic path.
 473                  * Resetting this prevents additional WARN() from panicking the
 474                  * system on this thread.  Other threads are blocked by the
 475                  * panic_mutex in panic().
 476                  */
 477                 panic_on_warn = 0;
 478                 panic("panic_on_warn set ...\n");
 479         }
 480
 481         print_modules();
 482         dump_stack();
 483         print_oops_end_marker();
 484         /* Just a warning, don't kill lockdep. */
 485         add_taint(taint, LOCKDEP_STILL_OK);
 486 }
 487
 488 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
 489 {
 490         struct slowpath_args args;
 491
 492         args.fmt = fmt;
 493         va_start(args.args, fmt);
 494         warn_slowpath_common(file, line, __builtin_return_address(0),
 495                              TAINT_WARN, &args);
 496         va_end(args.args);
 497 }
 498 EXPORT_SYMBOL(warn_slowpath_fmt);
 499
 500 void warn_slowpath_fmt_taint(const char *file, int line,
 501                              unsigned taint, const char *fmt, ...)
 502 {
 503         struct slowpath_args args;
 504
 505         args.fmt = fmt;
 506         va_start(args.args, fmt);
 507         warn_slowpath_common(file, line, __builtin_return_address(0),
 508                              taint, &args);
 509         va_end(args.args);
 510 }
 511 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
 512
 513 void warn_slowpath_null(const char *file, int line)
 514 {
 515         warn_slowpath_common(file, line, __builtin_return_address(0),
 516                              TAINT_WARN, NULL);
 517 }
 518 EXPORT_SYMBOL(warn_slowpath_null);
 519 #endif
 520
 521 #ifdef CONFIG_CC_STACKPROTECTOR
 522
 523 /*
 524  * Called when gcc's -fstack-protector feature is used, and
 525  * gcc detects corruption of the on-stack canary value
 526  */
 527 __visible void __stack_chk_fail(void)
 528 {
 529         panic("stack-protector: Kernel stack is corrupted in: %p\n",
 530                 __builtin_return_address(0));
 531 }
 532 EXPORT_SYMBOL(__stack_chk_fail);
 533
 534 #endif
 535
 536 core_param(panic, panic_timeout, int, 0644);
 537 core_param(pause_on_oops, pause_on_oops, int, 0644);
 538 core_param(panic_on_warn, panic_on_warn, int, 0644);
 539
 540 static int __init setup_crash_kexec_post_notifiers(char *s)
 541 {
 542         crash_kexec_post_notifiers = true;
 543         return 0;
 544 }
 545 early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
 546
 547 static int __init oops_setup(char *s)
 548 {
 549         if (!s)
 550                 return -EINVAL;
 551         if (!strcmp(s, "panic"))
 552                 panic_on_oops = 1;
 553         return 0;
 554 }
 555 early_param("oops", oops_setup);